摘要: 在最小二乘支持向量机的模型选择问题中,基于再抽样技术的模型选择方法,不能从根本上解决计算强度过高的问题。为此,提出基于模型复杂程度进行惩罚的新方法——秩准则,给出估计最小二乘支持向量机调谐参数的快速稳健算法。实例研究表明,该方法不仅能保证模型的预测精度和稳健性,而且在计算速度上优于快速Bootstrap方法。
关键词:
最小二乘支持向量机,
模型选择,
Bootstrap方法,
惩罚方法,
秩准则
Abstract: For Least Squares-Support Vector Machines(LS-SVM), model selection techniques based on re-sampling strategy suffer from high computational burden. To solve this problem, this paper proposes a new model selection method——order criterion, based on a penalization of the model complexity. A new fast robust algorithm is given for estimating the tuning parameter. Simulation study shows the proposed method is the huge merit over the fast bootstrap strategy on computational time, at the same time it is close to the Fast Bootstrap(FB) on both model prediction accuracy and model robustness.
Key words:
Least Squares-Support Vector Machines(LS-SVM),
model selection,
Bootstrap method,
penalization method,
order criterion
中图分类号:
陈建东, 王小明. LS-SVM模型选择的秩准则及其比较[J]. 计算机工程, 2011, 37(18): 185-187.
CHEN Jian-Dong, WANG Xiao-Meng. Order Criterion of LS-SVM Model Selection and Its Comparison[J]. Computer Engineering, 2011, 37(18): 185-187.